Alt Benchmark
#1 Holiday 2026: read-only discovery tasks
Top retail and travel sites (public pages only), 5 agents from several vendors, 4 discovery and qualification tasks each.
Publishes Mid-November 2026Methodology
No results yet.
Results publish on Mid-November 2026 with every run replayable. We do not preview numbers before the methodology is complete.
What each result will show
| Field | Example |
|---|---|
| Task | Answer: is cancellation free until 48 h before arrival? |
| Agents | Claude CU · Gemini CU · Browser Use · Stagehand · Skyvern |
| Outcome per agent | Completed / stalled at step N / blocked |
| Evidence | Recording, DOM hash, model version, date |
| Failure class | e.g. policy cannot resolve to yes/no |
Site owners get their own results privately before publication, and can opt out.